A model and multi-core parallel co-evolution algorithm for identifying cancer driver pathways

The identification of cancer driver pathways has become a significant issue in the field of bioinformatics in recent years, for driver pathways are crucial for understanding the pathology of cancer. In the current proposed identification methods, three common problems are of concern. (1) Strong restriction of mutual exclusivity is generally demanded for each individual sample. (2) Parameters are needed to be preset to balance the contributions of various omics data. (3) Only the interactions among genes within a gene set are applied to evaluate subnet importance. Therefore, a new driver pathway identification method is proposed in this study. The maximum weight submatrix model is first formulated based on Coverage, Mutual exclusivity and Subnet importance (CMS). A harmonic mean based measurement is devised to relax the constraint of mutual exclusivity. In addition, besides the gene interactions within a gene set, the interactions among genes inside and outside the gene set are also adopted to evaluate the subnet importance. Then a Multi-core Parallel Co-evolutionary Algorithm (MPCA) is designed for solving the CMS model (MPCA-CMS). Compared to the gene sets discovered using other cutting-edge methods, those recognized by the MPCA-CMS method not only contain more genes enriched in known signaling pathways, but also have greater performance in terms of subnet importance. Simultaneously, the higher efficiency of the MPCA-CMS method makes it more convenient to be applied in practical research. The source code are available at https://github.com/CXiaorong/MPCA_CMS. •Introduce a new measure of mutual exclusivity based on harmonic mean.•Devise a new approach for estimating subnet importance.•Present a parameter-free maximum weight submatrix identification model CMS.•Design a multi-core parallel co-evolution algorithm to solve the CMS model.•Conduct extensive trials using both real and simulated cancer datasets.